Part 27: One Word, Two Boys: Suki, and the Retranslation That Proved Nothing Is Neutral

In 2019, Netflix acquired Neon Genesis Evangelion and put it in front of a global audience with a new English dub and new subtitles. The translation work was supervised by Dan Kanemitsu. For a great many people it was the first legal, convenient way to watch the most important television anime ever made.

The fandom set itself on fire within about a day, over one word.

In episode 24, Kaworu Nagisa says something to Shinji Ikari. The Japanese word is suki. The older English version — the one a generation grew up on — rendered it, as widely remembered, along the lines of "I love you." The 2019 version rendered it "I like you."

I'm reporting the wordings as they circulated, because this argument has been re-litigated so many times that the quotations have drifted; the shape is what matters and the shape is not in dispute. Kanemitsu defended the choice publicly, on grounds of literalism and preserved ambiguity — the position being, roughly, that the Japanese is not as declarative as the old English made it, and the English should not decide what the Japanese left open.

That defence is correct. It is also, I'm going to argue, one of the more instructive wrong answers in the history of this craft.

Start with the fact that both are wrong

Suki does not mean "I love you." It also does not mean "I like you."

It covers a range English splits across at least two words and a great deal of social context. You can suki a person you intend to marry. You can suki takoyaki. Japanese keeps a heavier word, aishiteru, for the freighted declaration, and it is used sparingly enough that its absence here is genuinely meaningful — Kaworu did not reach for the big word. So the older translation was doing something: it was resolving an ambiguity in the direction of the maximum, and that is a real liberty.

But look at what "like" does in English. English "like" is not neutral, not ambiguous, and not a midpoint. It is specifically and actively casual. In English, "I like you" is what you say when you are declining to say the other thing. It carries a definite negative charge against romantic weight; it is the word people use to let someone down, or to stay safe. It does not open a space. It closes one.

So: the Japanese word is a wide range. Candidate A resolves it to the top of the range. Candidate B resolves it to the bottom of the range and pretends that isn't a resolution.

This is Part 7's honorifics and Part 14's pronouns wearing yet another coat. English has no slot. The gap is structural, not semantic. Nobody misunderstood the Japanese — everyone involved knew exactly what suki spans — and understanding it perfectly gets you precisely nowhere, because the target language does not have the shape you need.

Literalism is not neutrality, and it is not a way out

Here's the move that makes this essay worth writing.

When the target language has no equivalent, the translator does not get to decline. There is no abstain option on the form. Something has to be printed on the screen and every available string is an interpretation. The choice is not between interpreting and not interpreting. It is between interpretations.

What literalism offers is not neutrality. It offers deniability. It lets a choice be made while presenting itself as the absence of a choice — I didn't decide anything, I just wrote what it said. But "like" is a decision. It has a direction. It landed somewhere. And in a scene between two boys, at the emotional apex of the loneliest character in the medium, it landed in the direction of less, which is not a random direction, and which a very large number of viewers experienced as the moment being quietly taken away from them.

I don't need to allege intent to make this point and I'm not going to. I'll take the stated reasoning at face value: someone was trying to be accurate. The lesson is more interesting if it was done in good faith, because then it demonstrates the actual thesis — you cannot get out of the politics of a choice by being literal. The literal option is also a choice. It just gets to call itself the floor.

Fidelity to what?

And now the deeper problem, which Part 8 saw coming when it argued for the virtue of bad English.

A line is not the only thing in a scene. There is framing, and music, and silence, and everything the show spent twenty-three episodes building. The scene around suki is doing enormous work. It is built as an apex. It is shot and scored as an apex. Shinji has been starved of exactly this for the entire series, and the show knows it, and the audience knows it.

Drop "I like you" into that architecture and the English word actively contradicts the scene it is in. Everything visual and musical says this is the largest thing that has ever happened to this boy, and the text says casual, mild, friendly. That is not preserved ambiguity. That is a collision, and the viewer resolves the collision by concluding that something has been withheld from them.

So the more literal version is the less faithful version. Not because the lexicographer was wrong, but because fidelity is a property of the scene, not the sentence — and a translation can be word-accurate and scene-false at the same time. Part 3 found the same thing from the other direction: Nausicaä's title crossed intact while the film was butchered around it. Accuracy at the level of the string is worth very little if the thing the string is embedded in has been broken.

The other thing that went missing, which nobody translated at all

There is a second loss in the 2019 release and it is the more instructive one, because no translator was involved in it whatsoever.

The original series ended every episode with a cover of Fly Me to the Moon. Not incidental — twenty-six episodes of the strangest, loneliest show on television resolving, every single time, into a soft standard from 1954. The tonal joke of it, the tenderness of it, the way it let you down off the ceiling week after week. It is, for a lot of people, inseparable from what the show is.

The Netflix release did not have it. The reporting at the time pointed to music licensing — the song is a Western standard with rights holders and a global streaming deal is not the same instrument as a 1995 Japanese TV broadcast. I'm hedging the specifics because the details were murky then and are murkier now.

What isn't murky: a piece of the work did not cross, and the reason had nothing to do with language, meaning, craft, fidelity, or any decision any translator made. A rights holder and a lawyer somewhere failed to reach a number.

And notice the asymmetry in how those two losses were received. One word — suki — generated weeks of furious argument, thinkpieces, a public defence from the translation supervisor. The missing ending theme generated a shrug and a few sad posts. Yet by any honest accounting the song is more of the work than the word is. Twenty-six occurrences. A structural part of the show's emotional architecture.

Why the difference? Because suki was a choice, made by a person, that could have gone otherwise — and Part 23 established the rule that governs this entire field: the visible failure gets punished, the invisible one gets a pass. A licensing failure has no author to argue with. There is no one to be wrong. It reads as weather.

Which is Part 18's sieve arriving in a new place. That essay found that the biggest determinant of what crosses is not craft but a purchasing decision. This is the same fact operating inside a single work: what you receive of Evangelion was shaped more by a music rights negotiation than by any translator's judgement about any line. The translator gets the argument. The lawyer gets the outcome.

What retranslation is actually for

The last thing, and it's why this belongs in this series rather than in a fandom postmortem.

Every retranslation is a confession.

As long as there is one English version, it can pass for the work. It is transparent; you look through it and believe you're seeing the thing. It is a window.

Put a second version beside it and the window becomes a pane of glass with fingerprints on it. Two English versions that disagree are proof, sitting right there, that somebody chose — that the first one was never a transcription, that a person sat in a room and decided, and could have decided otherwise. You cannot un-see it afterward. Every line in both versions is now visibly a decision.

This is the single healthiest thing that can happen to a translated work, and it is why the 2019 fight, for all its heat and all its bad faith on every side, was worth more than another decade of quiet consensus. For twenty years English speakers had a version and mostly mistook it for the show. Now they have two, and they know it's a show that has to be carried, and they know somebody's hands are on it.

Part 21 said the translator's success condition is invisibility. Retranslation is the mechanism that makes them visible — the only reliable one the field has. It's an audit. It's the fans doing to the translation exactly what a good editor would have done, twenty years late, at volume, on the internet, badly.

It is still an audit. It is still more than the machine in Part 24 will ever get, because a machine translation that nobody thinks of as authored is a translation nobody will ever think to retranslate.

The numbers

Suki reads Destiny 6, Heart 3, Personality 3. Love reads Destiny 9, Heart 11, Personality 7. Like reads Destiny 1, Heart 5, Personality 5.

Three words. Three completely disjoint readings. Not one number shared between any pair of them.

The engine says suki is neither "love" nor "like," and is not between them, and does not resemble either. Which is, as it happens, the exact thesis of this essay, arrived at by a machine that counts letters in a romanization of a Japanese word it cannot read, comparing it to two English words it also cannot read.

It's right. It's right the way a stopped clock is right, and I want to be precise about why, because "the engine agrees with me" is the most seductive sentence available to me and I have twenty-six parts of practice writing it. The engine produced three different numbers because s-u-k-i, l-o-v-e, and l-i-k-e are three different strings. It would have produced three different numbers for any three different words. It would have produced three different numbers for suki, suki, and sukiii. There is no finding here. There is a spelling, and a coincidence of shape between what the spelling did and what I wanted to say.

Kaworu Nagisa reads Personality 11. Neon Genesis Evangelion reads Heart 11. Netflix reads Personality 22, a master number, which I could tell you means something about a company that carried this show to more people than everyone before it combined and fumbled one word doing it.

I could tell you that. Somebody should stop me. In three parts, somebody does.